BIS Working Papers

No 1046
The case for convenience:
how CBDC design choices
impact monetary policy
pass-through
by Rodney Garratt, Jiaheng Yu and Haoxiang Zhu

Monetary and Economic Department

November 2022

JEL classification: E42, G21, G28, L11, L15.

Keywords: central bank digital currency, interest

on reserves, payment convenience, deposit rates,
bank lending
BIS Working Papers are written by members of the Monetary and Economic
Department of the Bank for International Settlements, and from time to time by other
economists, and are published by the Bank. The papers are on subjects of topical
interest and are technical in character. The views expressed in them are those of their
authors and not necessarily the views of the BIS.

This publication is available on the BIS website (www.bis.org).

© Bank for International Settlements 2022. All rights reserved. Brief excerpts may be
reproduced or translated provided the source is stated.

ISSN 1020-0959 (print)

ISSN 1682-7678 (online)
The Case for Convenience: How CBDC Design Choices
Impact Monetary Policy Pass-Through∗
Rodney Garratt† Jiaheng Yu‡ Haoxiang Zhu§

Abstract
Banks of different sizes respond differently to interest on reserves (IOR) policy. For
low IOR rates, large banks are non-responsive to IOR rate changes, leading to weak
pass-through of IOR rate changes to deposit rates. In these circumstances, a central
bank digital currency (CBDC) may be used to provide competitive pressure to drive up
deposit rates and improve monetary policy transmission. We explore the implications
of two design features: interest rate and convenience value. Increasing the CBDC
interest rate past a point where it becomes a binding floor, increases deposit rates
but leads to greater inequality of market shares in both deposit and lending markets
and can reduce the responsiveness of deposit rates to changes in the IOR rate. In
contrast, increasing convenience, from sufficiently high levels, increases deposit rates,
causes market shares to converge and can increase the responsiveness of deposit rates
to changes in the IOR rate.

Keywords: central bank digital currency, interest on reserves, payment convenience, deposit
rates, bank lending
JEL Classification: E42, G21, G28, L11, L15

∗
The views expressed are those of the authors and do not necessarily reflect those of the BIS. We thank
Todd Keister for multiple discussions in the early stages of this work. We also thank Yu An, David Andol-
fatto, Ben Bernanke, Darrell Duffie, Zhiguo He, Michael Kumhof, Jiaqi Li, Debbie Lucas, Monika Piazzesi,
Alessandro Rebucci, Daniel Sanches, Antoinette Schoar, Harald Uhlig, Yu Zhu, and Feng Zhu, as well as
seminar participants at the Bank of Canada, Luohan Academy, MIT Sloan, Nova SBE, Microstructure On-
line Seminars Asia Pacific, the University of Hong Kong, the Canadian Economics Association, the JHU
Carey Finance Conference, the Philadelphia Workshop on the Economics of Digital Currencies, the CB &
DC Online Seminar, Boston University, the Federal Reserve Bank of New York, Shanghai Advanced Institute
of Finance, MIT Computer Science and Artificial Intelligence Lab, the 2021 ECB Money Market Conference,
AFA 2022 and the Future(s) of Money Conference, Paris 2022.
†
Bank for International Settlements. Email: [email protected]
‡
MIT Sloan School of Management. Email: [email protected].
§
MIT Sloan School of Management and NBER. Email: [email protected]. Haoxiang Zhu’s work on this paper
was completed prior to December 10, 2021, when he joined the U.S. Securities and Exchange Commission.

1
 “If all a CBDC did was to substitute for cash – if it bore no interest and came
without any of the extra services we get with bank accounts – people would
probably still want to keep most of their money in commercial banks.”
—Ben Broadbent, Deputy Governor of the Bank of England, in a 2016 speech

1 Introduction

A central bank digital currency (CBDC) “is a digital payment instrument, denominated in
the national unit of account, that is a direct liability of the central bank” (BIS, 2020). Over
the last few years, interest in CBDC has grown to the point where at present 90 percent of
central banks are investigating options for introducing CBDCs (BIS, 2021). As indicated in
Broadbent’s remarks (in the epigraph), policymakers initially contemplated a CBDC that
duplicated features of cash, without adding design characteristics that would make it more
likely to compete with money issued by commercial banks – the so called disintermediation
problem. However, more recently central banks have taken a broader view, and have been
more open to the possibility that CBDCs can help them to fulfill their mandates, either in
the present or the future. Central banks are increasingly viewing CBDCs as a way to improve
the payment system, promote financial inclusion, enhance monetary policy transmission, and
reduce systemic risk (BIS, 2020).
The likelihood that a CBDC will achieve any of the desired central bank objectives
depends upon its design features and how they interact. CBDCs can offer both pecuniary
benefits, in the form of interest payments, and nonpecuniary benefits. These can include a
host of features that enhance the performance of CBDC as a medium of exchange. Examples
include the quality of the user interface, processing speed, privacy and access to markets.1
1
A publicly provided CBDC could be less expensive to use and more widely accessible than existing,
privately offered payment methods, and it could offer access to new platforms and services. See, for example,
the Bank of England discussion paper on CBDC (Bank of England, 2020) which describes the potential for
third-party payment interface providers to provide overlay services on top of CBDC balances. A CBDC
could also provide privacy. Policy makers have argued that the central bank is specially positioned to
provide privacy in payments because the central bank does not have a profit motive to exploit consumer

2
We lump these possibilities together under the heading “payment convenience.”
In this paper, we explore the implications of introducing a CBDC that is interest-bearing
and offers payment convenience. All else equal, consumers will prefer to hold a currency that
pays higher interest. However, if a currency is easier to use, or accepted at more places, then
these non-pecuniary benefits may offset interest payments. Hence both interest rate and
payment convenience are important choice parameters for a CBDC, as both will determine
consumer demand for CBDC and hence its ultimate impact on monetary policy objectives.
We seek to evaluate design choices for CBDC in a model that is descriptive of the current
US financial system, which is characterized by large excess reserves and in which the main
monetary policy variable is interest on reserves (IOR).2 Crucially, we also want our analysis
to capture heterogeneity in bank size. Bank size matters for two key reasons. First, bank
size impacts the cost basis of loans, as it determines the likelihood of retained reserves. Loan
issuance involves the creation of deposits. When these deposits are spent by the borrower
there is a chance that the recipient will belong to the same bank, in which case there is
no associated transfer of reserves to another bank. This likelihood is not negligible for the
largest banks in the US economy and hence the impact of retained reserves should not be
ignored.3 Second, large bank deposits may offer a higher convenience value than small bank
deposits. For example, a large bank could have a more expansive network of branches and
payment data (Lagarde, 2018).
2
In the United States, IOR has been paid since October 2008. Since the financial crisis of 2008-09, interest
on excess reserves (IOER) has become the Federal Reserve’s main policy tool to adjust interest rates. In
July 2021, the Federal Reserve renamed IOER to interest on reserve balances (IORB), as required reserves
are currently zero. For simplicity, we use the acronym IOR.
3
During 2010-2020, based on Call Reports data, the top four largest banks captures 35% of US deposit
market. Their deposit market shares are large and stable over the last decade, with the averages listed in the
following: Bank of America (11%), Chase (10%), Wells Fargo (10%), and Citi (4%). Given the stationarity
of deposit shares, and the fact that deposits are created through commercial banks making loans (Bank of
England Quarterly Bulletin, 2014), the fraction of a lent dollar that ultimately flows to each bank should
approximate the deposit shares, regardless of the number of transactions or transfers. Hence the high deposit
market shares of the largest banks have non-negligible impact on their opportunity costs of making loans.
In local deposit markets, concentration is also salient. Drechsler, Savov, and Schnabl (2017) measure this
concentration by Herfindahl index (HHI), calculated by summing up the squared deposit-market shares of
all banks that operate in a given county. Their calculation indicates that around 50% of US counties have an
HHI that is higher than 0.3 (at least one bank’s deposit share is greater than 30%), and 25% of US counties
have an HHI that is higher than 0.5 (at least one bank’s deposit share is greater than 50%).

3
ATMs, a better mobile App, or a wider range of other (unmodeled) services. The higher
convenience value of its deposit may allow the large bank to offer a lower deposit rate than
the small bank and yet still maintain a larger market share.
In our model, the CBDC is offered through commercial banks. While CBDC balances
are the direct liability of the central bank, we envision that commercial banks will act as
the central bank’s agents to conduct KYC (Know Your Customer) and AML (Anti-Money
Laundering). This “tiered” design of CBDC is consistent with the recent pilot of E-Krona in
Sweden, the CBDC experiment in China, and the Banking for All Act in the United States.4
The heterogeneous-bank model we develop is new. The first part of our analysis seeks
to validate the model by demonstrating that it explains aspects of US deposit markets that
are not explained by existing, homogeneous-bank models of the US economy with large
reserves. In particular, we are able to explain the observed lack of interest-rate pass-through
in US deposit markets. Interest rate pass-through in the US economy is far from complete.
Drechsler, Savov, and Schnabl (2017) find that “[f]or every 100 bps increase in the Fed funds
rate, the spread between the Fed funds rate and the deposit rate increases by 54 bps.” Duffie
and Krishnamurthy (2016) document a sizable dispersion of a broad range of money market
interest rates, which widened as the Fed raised its interest on reserves. Our analysis shows
that the low correlation between movements in policy rates that determine the fed funds
rate and movements in deposit rates can be partly attributed to the differential impact these
changes have on banks of different sizes.
The intuition behind this result is as follows. The large bank’s ability to offer a lower
rate than the small bank may place them at a “corner solution” introduced by the fact that
deposit rates cannot go below zero. The large bank will set a deposit rate of zero, that is
non-responsive to changes in the policy rate, until that rate rises to a level where the zero
4
The Act argues that Digital Dollar Wallets should provide a number of auxiliary services including debit
cards, online account access, automatic bill-pay and mobile banking. These features (in particular mobile
banking which could give access to a variety of platforms that customers of a particular bank might otherwise
not have access to) could result in a CBDC with its own convenience value. Similar provisions are outlined
in the ECB’s digital euro report (2020).

4
lower bound on deposit rates is no longer binding. We will show that the zero lower bound
on deposit rates binds only when IOR rate is low. Hence for low levels of IOR we expect
deposit rates set by a large share of the banking sector to be non-responsive to changes in
IOR. For high levels of IOR, the lower bound is not binding and we expect all banks to adjust
deposit rates in response to changes in IOR (or the federal funds rate if this becomes the
relevant opportunity cost of lending). We demonstrate that both of these model predictions
seem to be true empirically.
In light of these observations regarding weak pass-through for low IOR rates, we examine
how outcomes in both the deposit market (deposit shares at the different banks and deposit
rates) and the lending market (loan volumes and rates) may be impacted by the choice
of design characteristics of a CBDC. First, we vary the CBDC interest rate. Increasing the
CBDC interest rate while holding the IOR rate and the CBDC convenience value fixed raises
the deposit rates of both banks, thus bringing their weighted average closer to the IOR rate.
Setting the CBDC interest rate equal to the interest rate on reserves would result in full
monetary policy pass-through. However, by forcing both banks to raise interest rates, a
higher CBDC interest rate makes it more difficult for the small bank to compete with the
large bank by offering higher deposit rate. Thus, a higher CBDC interest rate reduces the
market share of the small bank in deposit and lending markets, further widening the large
bank-small bank gap.
Second, we vary the CBDC convenience value, holding the IOR rate and CBDC interest
rate fixed. Making the CBDC more convenient weakens the market power of the large
bank by narrowing the convenience gap between the two banks. For example, by hosting a
convenient CBDC, a small community bank partially “catches up” with large global banks
in offering payment functionalities. The most immediate implication is that a convenient
CBDC results in a lower deposit rate at the small bank, because the small bank does not have
to compensate depositors as much for forgoing the large bank’s convenience. The deposit
rate at the large bank initially remains unchanged, and hence the average deposit rate for

5
the market falls as convenience is increased from zero. However, as convenience rises, a point
is eventually reached where the large bank is no longer constrained by the zero lower bound
and starts to raise its deposit rate to compete with the small bank. The result is an increase
in the average deposit rate. The implication is that for any given level of IOR, pass-through
of IOR to deposit rates is reduced for low levels of convenience and increased for high levels
of convenience.
Finally, we address the issue of how a given CBDC design impacts the sensitivity of
deposit rates to changes in the IOR rate. In the equilibrium where the lower bound on
deposit rates is not binding, pass-through is complete regardless of the levels of the CBDC
interest rate or convenience value. In equilibria where the lower bound is binding, we establish
two results that hold under some distributional assumptions. First, within the constrained
equilibrium, where the large bank’s deposit rate is non-responsive to changes in the IOR
rate, the response of the small bank’s deposit rate to increases in the IOR rate decreases
as the CBDC interest rate increases and increases as convenience increases. Second, higher
levels of the CBDC interest rate increase the range of IOR rates for which the large bank’s
deposit rate is non-responsive and higher levels of convenience decrease the range. Thus, a
positive interest on CBDC necessarily weakens monetary policy transmission from IOR to
deposit rates when the IOR rate is low, while increasing convenience necessarily increases
monetary policy transmission.
The paper is organized as follows. Section 2 provides a literature review. Section 3 intro-
duces the model and characterizes the constrained and unconstrained equilibrium. There is
a critical level of IOR rate at which the economy transitions from the the constrained to un-
constrained equilibrium. Hence, this analysis provides an explanation for weak pass-through
of IOR rate to deposit rates at low levels of IOR and strong pass-through at high levels.
Section 4 separately evaluates the direct impact on deposit rates of increasing the CBDC
interest rate s and convenience level v on deposit rates. Section 5 examines the sensitivity of
deposit rates to changes in the IOR rate under different CBDC designs. Section 6 concludes.

6
2 Literature

Our work builds on previous literature that has modelled deposit and lending markets in the
current regime of large excess reserves. In Martin, McAndrews, and Skeie (2016), a loan is
made if its return exceeds the marginal opportunity cost of reserves, which can be either the
federal funds rate or the IOR rate, depending on the regime. Our model differs in that we
have multiple banks and hence lent money may return to the same bank as new deposits.
Hence, our opportunity cost of lending is lower. Nevertheless, we share the conclusion that
the aggregate level of bank reserves does not determine the level of bank lending.
There is now a growing literature that seeks to examine the impact of CBDC on deposit
and lending markets. The conclusions vary and depend upon the level of competition, the
interest rate on the CBDC, and other features (e.g., liquidity properties of CBDC and reserve
requirements). Keister and Sanches (2021) consider a competitive banking environment in
which deposit rates are determined jointly by the transactions demand for deposits and the
supply of investment projects. If the CBDC serves as a substitute for bank deposits, then
its introduction causes deposit rates to rise, and the levels of deposits and bank lending to
fall.
In contrast, if banks have market power in the deposit market, the introduction of a
CBDC does not disintermediate banks, as banks can prevent consumers from holding the
CBDC by matching its interest rate. This lowers their profit margin, but does not lower the
level of deposits, and may even increase it. This is true in the model proposed by Andolfatto
(2021), where the bank is a monopolist. In that paper, an interest bearing CBDC causes
deposit rates to rise and the level of deposits to increase. Likewise, in that paper, banks have
monopoly power in the lending market, and, as in Martin, McAndrews, and Skeie (2016),
lending is not tied directly to the level of deposits, Hence, a CBDC does not impact the
interest rate on bank lending or the level of investment.
Chiu, Davoodalhosseini, Jiang, and Zhu (2019) also consider banks with market power
and show that an interest-bearing CBDC can lead to more, fewer or no change in deposits,

7
depending on the level of the CBDC interest rate. In an intermediate range of rates, the
CBDC impacts the deposit market in a manner similar to Andolfatto (2021) in that banks
offer higher deposit rates and increase deposits. Since, similar to Keister and Sanches (2021),
lending is tied to the level of deposits, adding the CBDC results in increased lending.
Our work is closest to Andolfatto (2021). We do not specify the overlapping generations
framework that he uses to make money essential. However, like Andolfatto (2021), in our
model, reserves are abundant, lending is determined by a performance threshold, and banks
have monopoly power in lending market. Hence, lending is determined not by deposit levels,
but instead by the opportunity cost of funds. In our model, this opportunity cost is lower
than the IOR rate, since we allow for the realistic feature that reserves come back to the
lending bank with a probability that depends on the deposit market share. Unlike Andolfatto
(2021), and the other works mentioned above, we incorporate two key design aspects of
CBDC, interest rate and convenience value, and we examine the combined impact these
features have on market outcomes in an environment with heterogeneous banks.
The impact of adding a CBDC can be richer in the presence of other frictions. In a
model with real goods and competitive banks, Piazzesi and Schneider (2020) find that the
introduction of CBDC is beneficial if all payments are made through deposits and the central
bank has a lower cost in offering deposits. However, they also find that the CBDC can be
harmful if the payer prefers to use a commercial bank credit line, but the receiver prefers
central bank money. Parlour, Rajan, and Walden (2022) argue that a wholesale CBDC
that enhances the efficiency of interbank settlement system could exacerbate the asymmetry
between banks if the CBDC does not distinguish net-paying and net-receiving banks. Agur,
Ari, and Dell’Ariccia (2022) consider an environment where households suffer disutility from
using a payment instrument that is not commonly used. They examine trade-offs faced by the
central bank in preserving variety in payment instruments and show that the adverse effects
of CBDC on financial intermediation are harder to overcome with a non-interest-bearing
CBDC.

8
 Fernández-Villaverde, Sanches, Schilling, and Uhlig (2021) extend the analysis of CBDC
to a Diamond and Dybvig (1983) environment in which banks are prone to bank runs. In
this setting, the fact that the central bank may offer more rigid deposit contracts allows it
to prevent runs. Since commercial banks cannot commit to the same contract, the central
bank becomes a deposit monopolist. Provided that the central bank does not exploit this
monopoly power, the first-best amount of maturity transformation in the economy is still
achieved.
Brunnermeier and Niepelt (2019) and Fernández-Villaverde, Sanches, Schilling, and Uhlig
(2021) derive conditions under which the addition of a CBDC does not affect equilibrium
outcomes. Key to their result is the central bank’s active role in providing funding to
commercial banks in order to neutralize the CBDC’s impact on their deposits.

3 Model and Equilibrium

3.1 Setup

The economy has a large bank (L) and a small bank (S).5 There are X = XS + XL reserves
in the banking system, where XS denotes the reserve holding of the small bank and XL
denotes the reserve holding of the large bank.6 For simplicity, the banks start off holding
reserves as their only asset, balanced by exactly the same amount of deposits. Following
Martin, McAndrews, and Skeie (2016), we assume that the level of reserves X is exogenously
determined by the central bank and is assumed to be large. The central bank pays the two
commercial banks an exogenously determined interest rate f on their reserve holdings, which
5
The assumption of two banks is, of course, a simplification. However, the situation may accurately
describe the retail depositors’ decision making process. Using survey data, Honka, Hortaçsu, and Vitorino
(2017) find that US consumers were, on average, aware of only 6.8 banks and considered 2.5 banks when
shopping for a new bank account. More than 80% considered fewer than 3 banks when shopping for a new
bank account.
6
We normalize the size of an individual loan to be $1, so reserves are in units of the standard loan size. For
example, if a loan size is $1 million and the actual reserve is $1 trillion, then in our model, X is interpreted
as $1 trillion/$1 million = 106 .

9
is called interest on reserves (IOR). The large and small banks pay depositors endogenously
determined deposit rates rL and rS , respectively. Thus, if nothing else happens, bank j’s
total profit would be Xj (f − rj ).
Commercial bank deposits are valuable not just for the interest they pay, but also for the
payment services they provide, the benefit of which we refer to as convenience value. The
convenience value of deposits in the small bank is normalized to be zero. The convenience
value of deposits in the large bank is a random variable δ ≥ 0 that has the twice differentiable,
cumulative distribution function G. We make the following assumption on G that we impose
throughout the paper:

Assumption 1. The function G satisfies −G0 (δ)/f < G00 (δ) < G0 (δ)/f for any δ ∈ [0, f −
s + v].

This condition ensures the second order condition of a bank’s optimization problem is
satisfied. The condition is also used in the comparative statics analysis. Bounding the
curvature of G bounds the masses of agents who value large bank’s deposits highly and lowly,
and ensures that both banks will compete to win additional depositors by raising their deposit
rates when f and other parameters, which we introduce in the following paragraphs, change.
Each depositor in the economy draws their large-bank convenience value δ independently
from the distribution G. This process reflects the idea that enhanced payment services are
not valued the same by all depositors.
The central bank offers a “retail” CBDC that is universally available. The CBDC has
two features: it pays an interest rate of s to depositors who use it and it provides a per-dollar
convenience value v ≥ 0 to users that is the same across all depositors.7 The convenience
value can be interpreted as a benefit that depositors receive from transacting using central
bank money. This benefit can include access to platforms on which CBDC can be spent,
7
The uniform nature of CBDC convenience value reflects the idea that CBDC should ideally create no
discrimination. This is also without loss of generality. If v varies across people, let’s say v = v̄ + ṽ, where v̄
is the average convenience value that can be adjusted by the central bank, and ṽ represents the individual
deviation from the average, we would only need to let G describe the distribution of δ − ṽ.

10
aspects of the mobile user interface (app features) or any other account services that are
associated with central bank accounts.
We assume that the CBDC is offered via commercial banks, and that money can be
transferred seamlessly between a depositor’s deposit account at a commercial bank and their
CBDC account offered via the commercial bank.8 Because a depositor can transfer money
between her deposit account and her CBDC account at no cost, she can obtain a convenience
value in payments that is equal to the maximum of the two options. A depositor at the
large bank receives convenience value max(δ, v) and a depositor at the small bank receives
convenience value max(0, v) = v. The convenience value v acts as a lower bound on the
payment convenience obtained by all depositors and thus narrows the gap between payment
convenience levels that depositors receive across banks of different sizes.
There is a unit mass of agents, and each potentially plays three roles: entrepreneur (bor-
rower), worker, and depositor. The main heterogeneity among the agents is their convenience
value for large bank deposits.
The model has four periods. At t = 0, the commercial banks set the deposit rates rL
and rS . The central bank sets the interest on reserves rate f , the CBDC interest rate s, and
the CBDC convenience value v. In the model, f, s, and v are exogenous, and rL and rS are
endogenous. At the start of the model, a fraction mL of agents have existing deposits at the
large bank and a fraction mS = 1 − mL of agents have existing deposits at the small bank.
The amount of deposits per capita across agents is identical. This means mL = XL /X and
mS = XS /X. Because users can seamlessly transfer money between the CBDC account and
the deposit account, a CBDC account offered via a commercial bank effectively provides the
same convenience value as the deposit account of that commercial bank. For this reason,
the CBDC interest rate s is a lower bound on banks’ deposit rates, i.e., rL ≥ s and rS ≥ s.
8
The Chinese CBDC experiment pivots around the e-CNY wallet mobile phone app. Embeded in the
app are interfaces connecting to deposit accounts at eight authorized commercial banks, as well as AliPay
and TenPay. Users can transfer money seamlessly between the deposit accounts and the e-CNY wallet, with
just a click, and make payments from the e-CNY wallet. The transfer incurs no fee. The CBDC launched
in Nigeria through the eNaira wallet app has the same characteristics. Chiu et al. (2019) also assume that
CBDC and deposits are perfect substitutes in terms of payment functions.

11
 At t = 1, the agents act as entrepreneurs and workers. Each agent is endowed with
a project, and each project requires $1 of investment and pays A > 1 with probability qi
and zero with probability of 1 − qi , where qi has the distribution function Q and A is a
commonly known constant. The expected payoff per dollar invested is thus qi A. Each agent
can only borrow from the bank where she keeps her deposit (the “relationship” bank). The
bank prices the loan as a monopolist. If the loan is granted, the entrepreneur pays $1 to
a randomly selected agent from the same population. The selected agent plays the role as
a worker and completes the project. The main point of introducing workers is to generate
some money flow in the economy.
At t = 2, agents play the role as depositors. Workers who receive wages choose where
to deposit the wage. The depositor can pick either the large bank or the small bank, and
within a bank, the depositor can pick either the bank’s own deposit account or the CBDC
account. These choices are made after considering the depositor’s own convenience value for
large bank deposits, the convenience value of the CBDC, and all relevant interest rates.
At t = 3, the projects succeed or fail. The banks earn interest on reserves and pays
depositors according to their deposit holdings and the deposit rates.

3.2 Bank deposit creation

For the purpose of illustration it is convenient to illustrate the deposit creation process by
considering a discrete set-up, in which we characterize the bank’s decision to make a single
loan. The condition on bank lending that we derive will be applicable to the continuum
model in which borrowers (i.e., the entrepreneurs) are infinitesimal.
The tables below show the sequence of changes in the large bank’s balance sheet in the
loan process. The changes in the small bank’s balance sheet in the loan process are entirely
analogous.

1. Before lending, the large bank starts with XL reserves. Its balance sheet looks like:

12
 Asset Liability

Reserves XL Deposits XL

2. If the large bank makes a loan of $1, it immediately creates deposit of $1 in the name
of the entrepreneur. The balance sheet of the bank becomes:

Asset Liability

Reserves XL Deposits XL
Loans 1 New Deposits 1

3. Eventually, the entrepreneur will spend her money to pay a worker. The large bank
anticipates that, in expectation, a fraction αS of the $1 new deposit will be transferred
to the small bank, leading to a reduction of reserves by the same amount. The fraction
αL remains in the bank because the worker has an account with the same bank. The
bank’s balance sheet becomes:

Asset Liability

Reserves XL − αS Deposits XL
Loans 1 New Deposits αL

If the large bank makes the $1 loan to entrepreneur i, and charges interest rate Ri , its
total expected profit, by counting all items in the balance sheet, will be

(X − α )f + [qi (1 + Ri ) − 1] − (XL + αL )rL . (1)

| L {z S } | {z } | {z }
Interest on reserves Gross profit on the loan Cost of deposits

If the large bank does not make the loan, then its total profit will be

XL (f − rL ). (2)

13
The large bank’s marginal profit from making the loan, compared to not making it, is

πi = qi (1 + Ri ) − (1 + f ) + αL (f − rL ). (3)
| {z } | {z }
Net profit on the loan Profit on deposit

In the expression of πi , the net profit on the loan reflects the true opportunity cost of
capital. Besides the usual profit on the loan, the large bank makes an additional profit equal
to αL (f − rL ). This is because each $1 lent out stays with the large bank with probability
αL and earns the bank the IOR-deposit spread of f − rL . The corresponding term for the
small bank’s marginal profit of lending is αS (f − rS ). In the equilibrium we characterize, it
will be the case that αL (f − rL ) > αS (f − rS ), i.e., the large bank’s convenience value of
deposits translates into an advantage in the lending market. Such a feature would not be
present if banks were homogeneous.

3.3 Equilibrium

We solve the model backward in time.

Deposit market at t = 2. A depositor with a large-bank convenience value of δ faces

four choices:

Large bank Small bank

Deposit CBDC Deposit CBDC

Convenience value max(δ, v) max(δ, v) v v

Interest rate rL s rS s

Obviously, the small bank attracts no depositors if rS < rL . So rS ≥ rL in equilib-

rium. For technical simplicity, whenever a depositor is indifferent between two choices, their
preference is the small bank, the large bank, and finally the CBDC, in this order.9
9
The tie-breaking rule between commercial banks and the CBDC is without loss of generality because
a commercial bank can always offer  above s so that depositors strictly prefer commercial bank deposits
to the CBDC. The tie-breaking rule also preserves continuity in the fractions of depositors as parameters
change to make depositors indifferent.

14
 We will characterize parameter conditions under which rS > rL . This implies that a
depositor with convenience value δ chooses the large bank if and only if

δ > v and rL + δ > rS + v ⇒ δ > rS − rL + v. (4)

Therefore, the eventual market shares of the banks in the newly created deposits are

αL = 1 − G(rS − rL + v) (5)

αS = G(rS − rL + v). (6)

Loan market at t = 1. In the previous section we derived the marginal profit of a bank
from making a loan. While the entrepreneur is infinitesimal here, expression (3) still applies.
The monopolist position of each bank in the lending market implies that a bank can make
a take-it-or-leave-it offer to the entrepreneur. The bank’s optimal interest rate quote would
be Ri = A − 1 (or just tiny amount below), and the entrepreneur, who has no alternative
source of funds, would accept. The lending bank takes the full surplus.
Hence, the large bank makes the loan if and only if

qi A − (1 + f ) + αL (f − rL ) > 0, (7)

or
1 + f − αL (f − rL )
qi > qL∗ = . (8)
A

Exactly the same calculation for the small bank yields the comparable investment threshold

1 + f − αS (f − rS )
qS∗ = . (9)
A

Choice of deposit rates at t = 0. Again, we start with the large bank. The large bank
makes profits in two ways. Because the large bank is a monopolist when lending to its

15
 R
customers, its first source of profit is on the loans, mL ql∗
(qA − 1 − f )dQ(q). The second
source of the large bank’s profit is on the interest rate spread. The existing deposit in the
banking system is X = XL + XS . As discussed above, the lending process also creates
new deposits. The amount of new deposit created by the large bank is mL (1 − Q(qL∗ )), by
the normalization that each loan is of $1. Likewise, the small bank creates new deposit
mS (1 − Q(qS∗ )). When the two banks compete for depositors by setting the deposit rates rL
and rS , we already show above that a fraction αL = 1 − G(rS − rL + v) of total deposits
end up with the large bank, enabling the large bank to collect a spread of f − rL per unit of
deposit held.
Adding up the two components, we can write the large bank’s total profit as

Z 1
ΠL = mL (qA − 1 − f )dQ(q) + [XL + XS + mL (1 − Q(qL∗ )) + mS (1 − Q(qS∗ ))]αL (f − rL )
∗
qL
Z 1
= mL [qA − (1 + f ) + αL (f − rL )]dQ(q) + [XL + XS + mS (1 − Q(qS∗ ))]αL (f − rL ).
∗
qL

(10)

Likewise, the small bank’s total profit is

Z 1
ΠS = mS [qA − (1 + f ) + αS (f − rS )]dQ(q) + [XL + XS + mL (1 − Q(qL∗ ))]αS (f − rS ).
∗
qS

(11)

As discussed before, the CBDC interest rate puts a lower bound on commercial banks’
deposit rates, i.e., rL ≥ s, and rS ≥ s. There are two cases. The first is that rL > s, so that
the CBDC interest rate does not constrain the commercial banks’ deposit rates. We call
the first case the unconstrained equilibrium. The second case is that rL = s, i.e., the CBDC
interest rate binds the large bank’s deposit rate. We call the second case the constrained
equilibrium.

16
Unconstrained equilibrium. Assuming that ΠL is strictly quasi-concave in rL , the suf-
ficient condition for a unique maximum of the function ΠL with respect to rL is

dΠL d[αL (f − rL )] dq ∗
= mL (1 − Q(qL∗ )) − mL [qL∗ A − (1 + f ) + αL (f − rL )] L
drL drL | {z } drL
=0
d[αL (f − rL )] dq ∗
+ [XL + XS + mS (1 − Q(qS∗ ))] − mS αL (f − rL )Q0 (qS∗ ) S
drL drL
= [XL + XS + mL (1 − Q(qL∗ )) + mS (1 − Q(qS∗ ))] · [(f − rL )G0 (rS − rL + v) − 1 + G(rS − rL + v)]
(f − rS )G0 (rS − rL + v)
− mS αL (f − rL )Q0 (qS∗ ) . (12)
A

Likewise, the first-order condition of the small bank is

dΠS
= [XL + XS + mL (1 − Q(qL∗ )) + mS (1 − Q(qS∗ ))] · [(f − rS )G0 (rS − rL + v) − G(rS − rL + v)]
drS
(f − rL )G0 (rS − rL + v)
− mL αS (f − rS )Q0 (qL∗ ) . (13)
A

For simplicity, let Q(·) be the uniform distribution on [0, 1]. And further impose a
stationarity condition that the market shares of deposits {αj } are identical to the starting
market shares {mj }. The first-order conditions simplify to

dΠL
0= = [X + αL (1 − qL∗ ) + αS (1 − qS∗ )] · [(f − rL )G0 (rS − rL + v) − 1 + G(rS − rL + v)]
drL
1
− αS αL (f − rL )(f − rS )G0 (rS − rL + v), (14)
A
dΠS
0= = [X + αL (1 − qL∗ ) + αS (1 − qS∗ )] · [(f − rS )G0 (rS − rL + v) − G(rS − rL + v)]
drS
1
− αL αS (f − rL )(f − rS )G0 (rS − rL + v). (15)
A

From the above conditions we derive

(rS − rL )G0 (rS − rL + v) + 2G(rS − rL + v) = 1. (16)

17
Proposition 1. Suppose that the profit function Πj is quasi-concave in rj , j ∈ {L, S} and
that G(v) < 0.5. Let rL and rS solve equations (14)–(15). If rL > s and rS > s, then it
is an unconstrained equilibrium that the banks set rL and rS as their deposit rates. In this
equilibrium:

1. The large bank sets a lower deposit rate (rL < rS < f ) and has a larger market share
(αL > αS ) than the small bank.

2. The large bank uses a looser lending standard than the small bank does (qL∗ < qS∗ ).

Proofs are in Appendix A.

The condition G(v) < 0.5 ensures that the CBDC does not increase the market share of
the small bank so much that it fully eliminates the large bank’s convenience value advantage
in its deposits. Consequently, the small bank still needs to compete by offering a higher
deposit rate than the large bank.
Further intuition of the equilibrium may be gained by considering an example. Suppose
that G(δ) = δ/∆, where δ ∈ [0, ∆] for a sufficiently large ∆. Then G0 (·) = 1/∆. The two
first-order conditions reduce to

f − rL rS − rL + v
=1− +B (17)
∆ ∆
f − rS rS − rL + v
= +B (18)
∆ ∆

where
1
α α (f
A∆ L S
− rL )(f − rS )
B≡ > 0. (19)
X + αL (1 − qL∗ ) + αS (1 − qS∗ )

As the total reserve X becomes large, B becomes close to zero. So the equilibrium deposit
rates of the two banks become approximately rL ≈ f − 23 ∆ + 13 v and rS ≈ f − 31 ∆ − 13 v. This
shows directly how an increase in convenience reduces the spread between deposit rates.

18
Constrained equilibrium. The second case of the equilibrium is that the CBDC interest
rate s becomes binding for the large bank. Recall the tie-breaking rule that at rL = s,
depositors use the large bank.
The small bank’s profit function and first-order condition are as before:

dΠS
0= = [X + αL (1 − qL∗ ) + αS (1 − qS∗ )] · [(f − rS )G0 (rS − s + v) − G(rS − s + v)]
drS
1
− αL αS (f − s)(f − rS )G0 (rS − s + v). (20)
A

By contrast, the large bank’s first order condition takes an inequality because the con-
jectured optimal solution is at the left corner:

dΠL
0> = [X + αL (1 − qL∗ ) + αS (1 − qS∗ )] · [(f − s)G0 (rS − s + v) − 1 + G(rS − s + v)]
drL rL ↓s
1
− αS αL (f − s)(f − rS )G0 (rS − s + v). (21)
A

Proposition 2. Suppose that the profit function Πj is quasi-concave in rj , j ∈ {L, S}.

Suppose that v < v̄ for some v̄ that may depend on f and s. Let rS solve equation (20). If,
at rS , equation (21) also holds, then it is a constrained equilibrium that the large bank sets s
and the small bank sets rS as their deposit rates. In this equilibrium:

1. The large bank sets a lower deposit rate (s < rS ) and has a larger market share (αL >
αS ) than the small bank.

2. The large bank uses a looser lending standard than the small bank does (qL∗ < qS∗ ).

The condition that v cannot be too high guarantees that the small bank still wishes to
compete by offering a higher deposit rate. This is analogous to the restriction on G(v) in
Proposition 1.
While the model described in this section is stylized, it potentially explains an important
fact about the U.S. deposit market: deposit rates are below and only partially responsive

19
to the key policy rate (Federal Funds rate and IOR) set by the central bank. We discuss
in Appendix B how our model, under a certain parameterization and without the laborious
calibration to the data, already generates predicted deposit rates that are largely similar to
actual U.S. deposit rates from 1986 to 2021. We contend that this conformance provides
essential support for the validity of the model’s predictions regarding the introduction of a
CBDC.
Parlour, Rajan, and Walden (2022) also analyze asymmetries in the banking sector and
their consequences. In their model, a bank that is a net payer incurs an additional settlement
cost and hence reduces lending, compared to net-receiving bank. In this sense, the net payer
bank in their model looks like the small bank in ours. Despite similar predictions on lending,
the two models are driven by different mechanisms. In our model, there is no exogenous
cost associated with interbank settlement; rather, the main advantage of the large bank in
lending is a higher likelihood that a lent dollar stays with the large bank and earns interest
on reserves from the central bank. Moreover, the size of the large bank’s advantage depends
on the interest rate paid on reserves and the CBDC design, including its interest rate and
convenience value, as we see in the next section.
Using confidential FedWire transaction data, Li and Li (2021) calculate the volatility
of daily net payments as a fraction of daily gross payments for various banks. They find
that banks with higher payment volatility pay a higher deposit rate and have lower loan
volume growth, controlling for a set of observables. While our model does not have payment
volatility, our predictions are consistent with the negative cross-sectional correlation they
compute between the deposit rate and lending.

4 Impact of CBDC Interest Rate and Convenience Value

In this section, we discuss the consequences of varying the CBDC interest rate s or conve-
nience value v. The results are summarized in Propositions 3 and 4.

20
4.1 Impact of CBDC interest rate s

When the federal reserve introduced the overnight reverse repo program (ONRRP) as a
temporary facility to support its IOR policy, it began by testing the facility by varying the
ONRRP rate between 1 basis point and 10 basis points, while holding the IOR rate fixed at
25 basis points. Here we examine how market outcomes change as s varies from a rate of 0
to f , while holding f fixed.
We focus on the case where, given a fixed value of v, f is sufficiently low that the con-
strained equilibrium applies. This case is most relevant to the current economic environment
in the United States. In the unconstrained equilibrium, market outcomes are invariant to
the CBDC interest rate s by definition.
Before we provide a formal statement of the comparative statics, it is useful to illustrate
the impact of CBDC interest rate changes in an example. The top row of Figure 1 plots
the behavior in the deposit markets as the CBDC interest rate rises from 0 to f = 2%.
The charts are computed numerically using a uniform distribution for G and a zero CBDC
convenience value (v = 0). As we see in the top left plot, raising the CBDC interest rate
increases the deposit rates of both banks as well as the weighted average deposit rates. The
top right plot shows the corresponding changes in deposit market shares αj , j = L, S, which
are easily computed from (5) and (6). Since the large bank’s deposit rate rises faster than
the small bank’s, the large bank gains market share from the small bank. Intuitively, the
small bank competes with the large bank primarily by offering a higher deposit rate. As s
increases, the maximum spread f − s shrinks, limiting the small bank’s ability to compete
with its interest rate choice. Once the deposit rates are equal at f , the large bank obtains
the entire market share of depositors, given the higher convenience value of its deposits.
The bottom row of Figure 1 illustrates the impact raising the CBDC interest rate has on
the lending market. Raising the CBDC interest rate changes the incentives to make loans
via the expected profit on the interest rate spread, αj (f − rj ). Because, as shown above,
both αS and f − rS decrease in s, so does αS (f − rS ). Thus, the small bank’s loan quality

21
 1+f −αS (f −rS )
threshold, qS∗ = A
, increases in s, and its loan volume, αS (1 − qS∗ ), decreases in s.
1+f −αL (f −rL )
In this example, the large bank’s loan quality threshold, qL∗ = A
, increases in s,
and its loan volume αL (1−qL∗ ), also increases due to its larger market share. In this example,
the total loan volume declines in s.

0.02 1

0.8 Small bank

0.015
Large bank
Deposit interest rate

Market share
0.6

0.01

0.4

0.005
Small bank 0.2
Large bank
Weighted average
0 0
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02
CBDC interest rate CBDC interest rate

0.68 0.35

0.3
0.678
0.25
Loan quality threshold

Small bank
Large bank
Loan volume

0.676
0.2 Total

0.15
0.674

Small bank 0.1

0.672 Large bank
0.05

0.67 0
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02
CBDC interest rate CBDC interst rate

Figure 1: Impact of CBDC interest rate on deposit and lending markets. Parameters:
G(δ) = δ/0.035, A = 1.5, X = 10, f = 0.02, v = 0.

The following proposition characterizes the impact of CBDC interest rate s on the deposit
and lending markets in more general cases.

Proposition 3. For a sufficiently large X, increasing the CBDC interest rate in a con-
strained equilibrium has the following impact on the deposit and lending markets:

22
 As s increases
Large Small

Deposit rates rL and rS ↑ ↑

Deposit market shares αL and αS ↑ ↓
Weighted average deposit rate ↑

Loan quality thresholds qL∗ and qS∗ ↑ ↑

Loan volume αL (1 − Q(qL∗ )) and αS (1 − Q(qS∗ )) ↑ or ↓ ↓
Total loan volume, i.e., total deposit created ? (↓ if G00 ≤ 0)

Most of the qualitative aspects illustrated in Figure 1 are true generally, and are ana-
lytically proven in Proposition 3. The exceptions are that, in general, the large bank’s loan
volume may go up or down in s and when G00 > 0, we do not know what will happen to
total loan volume.10

4.2 Impact of CBDC convenience value v

A convenient CBDC reduces the large bank’s convenience advantage and hence has an impact
even if its interest rate is zero. We illustrate the impact of a convenient CBDC by considering
this polar case in Figure 2. The top row shows the outcomes in the deposit market. As v
rises, the inconvenience disadvantage of the small bank shrinks. As long as the large bank’s
deposit rate remains at the floor rate, the small bank can afford to lower its interest rate
and still capture a growing market share. Once v gets large enough, the large bank responds
by raising its interest rate; however, the small bank can still afford to continue lowering its
deposit rate for the same reason that the convenience gap between the two banks continues
to shrink. Throughout this process the large bank loses market share and the small bank
gains market share, albeit at a slower rate once the large bank is no longer constrained. The
10
An example illustrating the ambiguity in large bank loans volumes is seen by setting G(δ) = δ/0.035, A =
1.05, X = 10, f = 0.02, v = 0. Then, in the constrained equilibrium, the large bank’s loan volume first
increases and then decreases with s.

23
overall impact of increasing the CBDC convenience value is the convergence of the deposit
rates and market shares for the two banks.11

0.01 0.8
Small bank Small bank
Large bank Large bank
Weighted average 0.7
0.008
Deposit interest rate

0.6

Market share
0.006

0.5

0.004
0.4

0.002
0.3

0 0.2
0 0.005 0.01 0.015 0 0.005 0.01 0.015
CBDC convenience value CBDC convenience value

0.679 0.35
Small bank
0.678 Large bank
0.3
0.677
Loan quality threshold

0.676 0.25
Loan volume

0.675
0.2
0.674

0.673 0.15

0.672
0.1 Small bank
0.671 Large bank
Total
0.67 0.05
0 0.005 0.01 0.015 0 0.005 0.01 0.015
CBDC convenience value CBDC convenience value

Figure 2: Impact of CBDC convenience value on deposit and lending markets. Parameters:
G(δ) = δ/0.035, A = 1.5, X = 10, f = 0.02, s = 0. The equilibrium transitions from
constrained to unconstrained at v = 0.01.

The CBDC convenience value has a nuanced impact on the weighted average deposit
rates. In a constrained equilibrium, a higher v results in a lower weighted average deposit
rate when the large bank’s deposit rate is at the lower bound. That is, a convenient CBDC
weakens the transmission of monetary policy to the deposit market through IOR. Once the
11
In fact, a modest CBDC convenience value may be enough to fully level the playing field. Under the
uniform distribution of large-bank preference δ, when v rises to the point v = ∆/2, depositors with δ > ∆/2
strictly prefer the large bank, and depositors with δ < ∆/2 strictly prefer the small bank. That is, the
deposit market shares become equal and so do the deposit rates, loan quality thresholds, and loan volume.

24
economy transitions to an unconstrained equilibrium with a sufficiently high v, however, a
higher CBDC convenience value increases the average deposit rate, increasing the transmis-
sion of monetary policy.
The bottom row of Figure 2 shows the outcomes in the lending market. Because the
two deposit rates and the deposit market shares get closer to each other as v rises, it is
unsurprising that the loan quality thresholds and loan volume of the two banks are also
getting closer to each other. In this example, the total loan volume is almost invariant to v,
and the most salient effect is the reallocation of loans from the large bank to the small one.
Figure 3 below further demonstrates the potential impact of CBDC convenience value on
loan volume, using a different parametrization of f = 3% and s = 1.25%. These parameters
lead to a constrained equilibrium, with rL = s. In this example, the total lending volume
(left axis) is first decreasing in v and then increasing in v. The magnitude of the axes suggests
that the more salient action is, again, the shift of lending from the large bank to the small
one.

0.3204 0.3
Total
0.3202 Small bank
Large bank 0.25
0.32
Loan volume

0.3198
0.2

0.3196

0.15
0.3194

0.3192
0.1
0.319

0.3188 0.05
0 0.005 0.01 0.015
CBDC convenience value

Figure 3: CBDC convenience value and loan volume. Total loan volume in on left axis.
Loan volume of the two banks are on right axis. Parameters: G(δ) = δ/0.035, A = 1.5, X =
10, f = 0.03, s = 0.0125. All values of v in the depicted range correspond to a constrained
equilibrium.

The next proposition summarizes the comparative statics with respect to v.

25
Proposition 4. For a sufficiently large X, the impact of increasing v is given in the following
table:

As v increases Constrained Unconstrained

Large Small Large Small

Deposit rates rL and rS Flat(=s) ↓ ↑ ↓

Deposit market shares αL and αS ↓ ↑ ↓ ↑
Weighted average deposit rate ↑ or ↓ (↓ if G00 ≤ 0) ↑ or ↓ (↑ if G00 ≥ 0)

Loan quality thresholds qL∗ and qS∗ ↑ ↓ ↑ ↓

Loan volume αL (1 − qL∗ ) and αS (1 − qS∗ ) ↓ ↑ ↓ ↑
Total loan volume, i.e., total deposit created ↑ or ↓ ↑ or ↓ (↓ if G00 ≥ 0)
As in the previous subsection, most of the comparative static results that apply for uni-
form G are true more generally and are stated in Proposition 4. There are a few exceptions.
The impact on the weighted average interest rate, αS rS + αL rL , is ambiguous. An increase
in v shifts market share to the small bank and reduces the small bank’s deposit rate, but
the small bank has a higher deposit rate to start with, so the overall effect can go in either
direction. In the constrained equilibrium, a concave G means that relatively more depositors
have a weak (but still positive) preference for large bank’s deposits, so a higher v quickly
eliminates the large bank’s advantage. As a result, the small bank can afford to reduce its
deposit rate quickly, leading to a lower weighted average αS rS +αL rL .12 In the unconstrained
equilibrium, rL increases in v. A convex G means that relatively more depositors have a
strong (but still positive) preference for large bank’s deposits, so the large bank raises rL
aggressively compared to the reduction in rS , leading to a higher αS rS + αL rL .13 The ambi-
guity in total loan volume for the constrained case is illustrated by the numerical example
in Figure 3 where G is uniform and total loan volume first decreases and then increases in
12
When A = 1.5, X = 10, f = 0.02, s = 0, G = Gamma(5, 150) with mean 1/30, then 0 < G00 (rS −rL +v) <
0
G (rS − rL + v)/f , yet weighted average deposit rate first increases then decreases in v, in the constrained
equilibrium.
13
When A = 1.5, X = 10, f = 0.02, s = 0, G = Gamma(3, 200) with mean 3/200, then −G0 (rS − rL +
v)/f < G00 (rS − rL + v) < 0, yet weighted average deposit rate first increases then decreases in v, total loan
volume first decreases then increases in v, in the unconstrained equilibrium.

26
v. Total loan volume unambiguously decreases in v in the unconstrained equilibrium if G
is weakly convex. Intuitively, the weighted average deposit rate increases in v, so the IOR-
deposit rate spread is compressed, which discourages lending. When G is strictly concave,
however, the change in total loan volume resulting from an increase in v can be in either
direction.

5 Sensitivity

We now address the issue of how a given CBDC design impacts the sensitivity of deposit
rates to changes in the IOR rate f . In the unconstrained equilibrium, deposit rates of both
the large and the small bank move one-for-one with the IOR rate f . This means the pass-
through of f is perfect when the large bank is not constrained, that is, when the large bank
competes with the small bank on the deposit rate margin. In the constrained equilibrium,
the large bank’s deposit rate is capped at CBDC interest rate s, and only the small bank’s
deposit rate reacts to changes in f , hence the pass-through of f to the average deposit rate
is much weaker.
When the IOR rate f is low, deposit and lending markets are characterized by the
constrained equilibrium and when f is high they enter into the unconstrained equilibrium.
Let f ∗ denote the threshold value of the economy transitions from the constrained equilibrium
to the unconstrained equilibrium. Since the pass-through of f is vastly different between the
constrained equilibrium and the unconstrained equilibrium, it is important to understand
how the CBDC interest rate s and the convenience value v affect the cut-off value of IOR,

27
f ∗ , that separates the two equilibria. We solve for f ∗ , from the following FOCs.

0 = [X + αL (1 − qL∗ ) + αS (1 − qS∗ )] · [(f ∗ − rS )G0 (rS − s + v) − G(rS − s + v)]

1
− αL αS (f ∗ − s)(f ∗ − rS )G0 (rS − s + v).
A
0 = [X + αL (1 − qL∗ ) + αS (1 − qS∗ )] · [(f ∗ − s)G0 (rS − s + v) − 1 + G(rS − s + v)]
1
− αS αL (f ∗ − s)(f ∗ − rS )G0 (rS − s + v). (22)
A

The following proposition summarizes the results.

Proposition 5. For a sufficiently large X:

1. In the unconstrained equilibrium, rL and rS move one-for-one with f ;

drS G00 (δ)

2. In the constrained equilibrium, df
decreases with s and increases with v if G0 (δ)
is
increasing in δ; and

3. f ∗ increases with s. f ∗ decreases with v if G00 (δ) ≥ 0.

With abundant reserves, the spread between IOR and the deposit rate is the main factor
that determines profits for the banks. In the unconstrained equilibrium, the two banks
compete primarily for deposit market shares, and the market shares do not vary with f . As
a result, rL and rS move one-for-one with f .
Part 2 of Proposition 5 requires that G00 (δ)/G0 (δ) is an increasing function as a sufficient
condition.14 Such distributions have decreasing probability density functions and large mass
at small values. Under these assumptions, in the constrained equilibrium, higher s or lower
v decreases the sensitivity of the small bank’s deposit rate rS to changes in f . This occurs
under the stated convexity condition because when s is higher, or v is lower, the convenience
value for the large bank of the marginal consumer that is indifferent between choosing the
small bank and the large bank takes a lower value where the convexity of G is lower. This
14
This condition is satisfied by Gamma distributions with a shape parameter less than 1. The proposition
is also true for exponential distributions, where G00 (δ)/G0 (δ) is a non-zero constant.

28
is where rS needs to react less to offset the change introduced by f . Since rS becomes less
sensitive to f as s increases or v decreases, pass-through is decreased.
A higher s increases the cutoff value f ∗ , which means the Fed needs to set a higher IOR
to enter into the high pass-through region. Intuitively, there needs to be a large spread
between IOR and s in order to induce the large bank to compete with the small banks via
its deposit rate policy. A higher s necessarily increases f ∗ . A higher v reduces the large
bank’s competitive advantage, and forces it to compete with the small bank sooner; that is,
a higher v decreases the cutoff value f ∗ . The sufficient condition for this is that G is convex.
Convex G means relatively more depositors have a strong preference for the large bank’s
deposits, so the large bank competes sooner on the deposit rate margin to compensate for
the reduction in the convenience advantage.

6 Concluding Remarks

Payment convenience is a crucial aspect of CBDC design that may be more desirable than
interest rate policy. A highly convenient CBDC produces sufficient competitive pressure in
deposit markets to raise deposit rates for any given level of IOR and increases the responsive-
ness of deposit rates to IOR rate changes. Convenience also has favorable effects on market
composition by leveling the playing field. Interest rate policy is less desirable in the sense
that it may weaken the responsiveness of deposit rates to IOR rate changes and it increases
the inequality of market shares.
An interesting aspect of our analysis is that the provision of CBDC impacts equilibrium
outcomes even though the currency is not held in equilibrium. Hence there is no disinter-
mediation. This is also true in Chiu et al. (2019) and Garratt and Lee (2021), where the
option to use CBDC changes the equilibrium outcome even it is not exercised. An exception
is Keister and Sanches (2021), where the CBDC has specific liquidity benefits that leads to
its use. The idea that a central bank introduces a program to influence market rates by

29
increasing the bargaining power of lenders is not new. Early descriptions of the overnight
reverse repurchase agreement facility that the Federal Reserve Bank of New York began test-
ing in September 2013 indicated that “the option to invest in ON RRPs [Overnight Reverse
Repurchase Agreement Facility] also would provide bargaining power to investors in their
negotiations with borrowers in money markets, so even if actual ON RRP take-up is not
very large, such a facility would help provide a floor on short-term interest rates...” (Frost
et al., 2015).
The results of our paper could be extended in multiple directions. One possible extension
is to add short-term investment vehicles such as money market mutual funds and repurchase
agreements that typically pay higher interest rates than bank deposits but cannot be easily
used for processing payments. If the CBDC pays a sufficiently high interest rate, it is
possible that money would flow out of these short-term investment vehicles into the CBDC,
i.e., investors would earn returns from the Fed rather than short-term Treasury Bills. This
additional channel is unlikely to affect lending because money market investors do not make
loans. Another possibility is to consider heterogeneous CBDC interest rates paid to banks
of different sizes, which adds yet another degree of freedom in the central bank’s toolkit.
In particular, the central bank could use heterogenous CBDC interest rates to fine-tune the
competitive positions of large and small banks. These extensions are left for future research.

Appendix A: Proofs

Proof of Proposition 1

Taking the difference of the two FOCs, we have (rS −rL )G0 (rS −rL +v) = 1−2G(rS −rL +v). If
rS ≤ rL , then the left-hand side is non-positive but the right-hand side is 1−2G(rS −rL +v) ≥
1 − 2G(v) > 0, a contradiction. So rS > rL . This implies that G(rS − rL + v) < 0.5 in
equilibrium, i.e., αL > αS .

30
 Let
1
α α (f
A L S
− rL )(f − rS )G0 (rS − rL + v)
B≡ > 0. (23)
X + αL (1 − qL∗ ) + αS (1 − qS∗ )

The two FOCs are separately written as

(f − rL )G0 (rS − rL + v) = αL + B, (24)

(f − rS )G0 (rS − rL + v) = αS + B. (25)

So both rL and rS are below f . Take the ratio:

f − rL αL + B αS
= >1> . (26)
f − rS αS + B αL

Hence, (f − rL )αL > (f − rS )αS , and qL∗ < qS∗ .

Proof of Proposition 2

By the assumption that the function ΠS is well behaved to admit a unique global maximum,
the derivative dΠS /drS should be strictly decreasing in rS . To show that rS > s, it is
sufficient that the right-hand side of (20) is positive at rS = s, i.e.,

1
[X + αL (1 − qL∗ ) +αS (1 − qS∗ )][(f − s)G0 (v) − G(v)] − G(v)(1 − G(v))(f − s)2 G0 (v) > 0. (27)
A

Clearly, the above equation holds at v = 0. By continuity, it also holds if v is below a cutoff,
say v̄. If v ∈ [0, v̄), we have rS > s = rL .
1
α α (f −s)(f −rS )G0 (rS −s+v)
A L S
Let B ≡ X+αL (1−qL ∗ )+α (1−q ∗ )
S
> 0. The two FOCs are separately written as
S

(f − s)G0 (rS − s + v) < αL + B, (28)

(f − rS )G0 (rS − s + v) = αS + B. (29)

Since B > 0, G0 (rS − s + v) > 0 , we know rS < f . Take the difference, we have 0 <

31
(rS − s)G0 (rS − s + v) < αL − αS . That is, αL > αS . It follows that (f − s)αL > (f − rS )αS
and qL∗ < qS∗ .

Proof of Proposition 3

Since the large bank is constrained by the lower bound, its deposit rate rises in step with the
CBDC interest rate s. Meanwhile, the small bank adjusts its equilibrium deposit rate at a
slower pace, continuing to balance its ability to maintain depositors while its profit margin
shrinks. To see how rS is affected by s, let ΓS = dΠS /drS , ΓL = dΠL /drL , and start with
the expression
∂ΓS ∂ΓS drS
0= + . (30)
∂s ∂rS ds

Because ∂ΓS /∂rS < 0, a sufficient condition for drS /ds > 0 is ∂ΓS /∂s > 0. Writing total
loan volume as V = αL (1 − qL∗ ) + αS (1 − qS∗ ), we have

∂ΓS  αL 
= −αL [(f − rS )G0 (rS − s + v) − G(rS − s + v)]
∂s A
+ (X + V )[−(f − rS )G00 (rS − s + v) + G0 (rS − s + v)]
1 ∂
+ (f − rS ) [αL αS (f − s)G0 (rS − s + v)]. (31)
A ∂s

On any closed region of f and s, the first and third term are bounded, by G being twice-
differentiable. So if X is sufficiently large, the second term dominates. Under the assumption
that G00 (δ) < G0 (δ)/f for δ ∈ [0, f −s+v], we have −(f −rS )G00 (rS −s+v)+G0 (rS −s+v) > 0,
so a sufficiently large X would imply that ∂ΓS /∂s > 0, and so is drS /ds.
Next, we show that rS − s decreases in s. We have

∂ΓS ∂
=X [(f − rS )G0 (rS − s + v) − G(rS − s + v)]
∂rS ∂rS
∂
+ {[αL (1 − qL∗ ) + αS (1 − qS∗ )] · [(f − rS )G0 (rS − s + v) − G(rS − s + v)]}
∂rS
 
∂ 1 0
− αL αS (f − s)(f − rS )G (rS − s + v) . (32)
∂rS A

32
The second and the third term are bounded on any closed region of rS . The first term equals
X[(f − rS )G00 (rS − s + v) − 2G(rS − s + v)]. Hence,

drS ∂ΓS /∂s (f − rS )G00 (rS − s + v) − G0 (rS − s + v)

=− → , (33)
ds ∂ΓS /∂rS (f − rS )G00 (rS − s + v) − 2G0 (rS − s + v)

G0 (rS −s+v)
as X becomes sufficiently large. We also have d(rS −s)/ds = drS /ds−1 = (f −rS )G00 (rS −s+v)−2G0 (rS −s+v)
,
whose denomintor is negative under the condition that G00 (δ) < G0 (δ)/f Hence, d(rS −
s)/ds < 0. This implies that αS = G(rS − s + v) is decreasing in s, and αL is increasing in s.
The weighted average interest rate is αS rS + αL s. Its derivative with respect to s is

 
d(αS rS + αL s) dαS drS dαL drS
= rS + αS + s + αL = [(rS − s)G0 (rS − s + v) + αS ] − 1 + 1.
ds ds ds ds ds
(34)
drS G0 (rS −s)
By the calculation earlier, as X becomes large, ds
−1 → (f −rS )G00 (rS −s+v)−2G0 (rS −s+v)
>
f
− f +r S
, where the inequality follows from G00 (δ) < G0 (δ)/f for any δ ∈ [0, f − s + v]. So, as
X becomes large,

d(αS rS + αL s) f f
>1− [(rS − s)G0 (rS − s + v) + αS ] ≥ 1 − > 0, (35)
ds f + rS f + rS

where the second last inequality follows from the large bank’s FOC that, limX→∞ (f −
s)G0 (rS − s + v) + G(rS − s + v) ≤ 1.
Now we turn to loan market outcomes. Since αS decreases in s and rS increases in
s, αS (f − rS ) is decreasing in s and qS∗ is increasing in s. The small bank’s loan volume,
αS (1 − qS∗ ), is then decreasing in s.
For the large bank’s loan quality qL∗ , we have

dqL∗ 1h 0 drS i
= − G (rS − s + v)(1 − )(f − s) − 1 + G(rS − s + v) . (36)
ds A ds

drS
For the first term in the brackets, we know that G0 (rS − s + v)(1 − ds
)(f − s) < (f −

33
s)G0 (rS − s + v), since drS /ds > 0. Also, from the large bank’s optimality condition, as X is
sufficiently large, we know that (f − s)G0 (rS − s + v) − 1 + G(rS − s + v) ≤ 0. That means
dqL∗ /ds > 0 and qL∗ is increasing in s. However, the impact of s on the large bank’s loan
volume αL (1 − qL∗ ) is ambiguous.
The total loan volume is αL (1 − qL∗ ) + αS (1 − qS∗ ). Its derivative with respect to s is

 
1 drS 1 1 drS
[2αS (f − rS ) − 2αL (f − s)]G0 (rS − s + v) − 1 − αL2 − αS2 . (37)
A ds A A ds

While the first term is positive, the last two terms are negative. It is, however, possible to
show that this derivative is negative if G00 (δ) ≤ 0 and X is sufficiently large. As X becomes
large, the two first-order conditions imply that

lim (f − s)G0 (rS − s + v) − (1 − G(rS − s + v)) ≤ 0,

X→∞ | {z }
αL

lim (f − rS )G0 (rS − s + v) − G(rS − s + v) = 0. (38)

X→∞ | {z }
αS

Multiplying αL to the first equation and αS to the second equation, we have

lim αL (f − s)G0 (rS − s + v) − αL2 ≤ 0,

X→∞

lim αS (f − rS )G0 (rS − s + v) − αS2 = 0. (39)

X→∞

Plugging these in Equation (37), we have, as X becomes large,

 
d ∗ ∗ 1 2 2 drS 1 1 drS
lim (αL (1 − qL ) + αS (1 − qS )) ≤ (2αL − 2αS ) 1 − − αL2 − αS2
X→∞ ds A ds A A ds
    
1 drS drS
= 1−2 αL2 + − 2 αS2 , (40)
A ds ds

drS
Because ds
< 1, ( drdsS − 2)αS2 < 0. If G00 (δ) ≤ 0 and X is sufficiently large, we know from
drS drS
the expression of ds
above that ds
≥ 12 . That means (1 − 2 drdsS )αL2 ≤ 0 as well. So the total

34
loan is decreasing in s in the limit. Because the limit is strictly negative, it is also negative
for finite but large enough X.

Proof of Proposition 4

First we consider the unconstrained equilibrium and then the constrained one.

The unconstrained equilibrium

1
We know that rL < rS < f , and αS < 2
< αL . Let ΓS = dΠS /drS , ΓL = dΠL /drL . To
calculate how rL and rS are affected by v, we take derivative of ΓL and ΓS at the equilibrium
values and obtain

∂ΓL ∂ΓL drL ∂ΓL drS

0= + + , (41)
∂v ∂rL dv ∂rS dv
∂ΓS ∂ΓS drL ∂ΓS drS
0= + + . (42)
∂v ∂rL dv ∂rS dv

drL drS ∂ΓL ∂ΓL ∂ΓL

We solve for dv
and dv
from above equations. Denote Av = ∂v
, AL = ∂rL
, AS = ∂rS
,
∂ΓS ∂ΓS ∂ΓS
Bv = ∂v
, BL = ∂rL
, and BS = ∂rS
. Then we have,

drL AS Bv − BS Av
= (43)
dv AL BS − BL AS
drS BL Av − AL Bv
= (44)
dv AL BS − BL AS

When X is sufficiently large, Av is dominated by X[(f − rL )G00 (rS − rL + v) + G0 (rS −

rL + v)], so Av ≈ X[(f − rL )G00 (rS − rL + v) + G0 (rS − rL + v)]. Similarly, Bv ≈ X[(f −
rS )G00 (rS − rL + v) + G0 (rS − rL + v)], AL ≈ X[−(f − rL )G00 (rS − rL + v) − 2G0 (rS − rL + v)],
AS ≈ X[(f − rL )G00 (rS − rL + v) + G0 (rS − rL + v)], BL ≈ X[−(f − rS )G00 (rS − rL + v) +
drL
G0 (rS − rL + v)], and BS ≈ X[(f − rS )G00 (rS − rL + v) − 2G0 (rS − rL + v)]. Hence, dv
→
(f −rL )G00 (rS −rL +v)+G0 (rS −rL +v) drS (f −rS )G00 (rS −rL +v)−G0 (rS −rL +v)
(rS −rL )G00 (rS −rL +v)+3G0 (rS −rL +v)
, and dv
→ (rS −rL )G00 (rS −rL +v)+3G0 (rS −rL +v)
. Since −G0 (δ)/f <
G00 (δ) < G0 (δ)/f , (rS − rL )G00 (rS − rL + v) + 3G0 (rS − rL + v) is positive, and (f − rL )G00 (rS −

35
 drL
rL + v) + G0 (rS − rL + v) is positive, so dv
> 0. Also, (f − rS )G00 (rS − rL + v) − G0 (rS − rL + v)
drS
is negative, so dv
< 0. So rL is increasing and rS is decreasing in v.
For deposit market share αS = G(rS − rL + v), we take the difference of the two FOCs,
and have

(rS − rL )G0 (rS − rL + v) + 2G(rS − rL + v) = 1. (45)

Write y = rS − rL + v, and take derivative of the above equation with respect to v, then we
have

dy
[3G0 (y) + (rS − rL )G00 (y)] − G0 (y) = 0 (46)
dv

dy
Since −G0 (δ)/f < G00 (δ) < G0 (δ)/f , we know that 3G0 (y)+(rS −rL )G00 (y) > 0, hence dv
> 0.
So αS is increasing in v, and αL is decreasing in v.
The weighted average deposit rate is αS rS + αL rL = αS (rS − rL ) + rL . Its derivative
with respect to v is

d(αS rS + αL rL ) dαS d(rS − rL ) drL

= (rS − rL ) + αS + (47)
dv dv dv dv
dαS 1 d(rS − rL ) drL dαS 1 d(rL + rS )
> (rS − rL ) + + = (rS − rL ) + ,
dv 2 dv dv dv
|{z} 2 dv
>0

1
where the inequality follows from αS < 2
and rS − rL decreasing in v. As X becomes large,

d(rS + rL ) (2f − rL − rS )G00 (rS − rL + v)

→ (48)
dv (rS − rL )G00 (rS − rL + v) + 3G0 (rS − rL + v)

The denominator is positive as −G0 (δ)/f < G00 (δ) < G0 (δ)/f . As rL + rS < 2f , when
d(rL +rS )
G00 (δ) ≥ 0, we have dv
≥ 0, and hence αS rS + αL rL increases in v.
For loan quality thresholds, since αL is decreasing in v and rL is increasing in v, qL∗ is
increasing in v. Since αS is increasing in v and rS is decreasing in v, qS∗ decreasing in v.

36
 For loan volumes, αL (1−qL∗ ) is decreasing in v, since αL is decreasing and qL∗ is increasing.
Similarly, αS (1 − qS∗ ) is increasing in v.
1+f α2L (f −rL )+α2S (f −rS )
Total loan volume equals αL (1 − qL∗ ) + αS (1 − qS∗ ) = 1 − A
+ A
. Its
derivative with respect to v is

 
1 dαS drL drS
[2αS (f − rS ) − 2αL (f − rL )] − αL2 − αS2 . (49)
A dv dv dv

dαS drL drS

where 2αS (f − rS ) − 2αL (f − rL ) < 0, dv
> 0, dv
> 0, and dv
< 0. We know −αL2 drdvL −
αS2 drdvS < −αS2 d(rLdv+rS ) . If G00 (δ) ≥ 0, we know from above that d(rL +rS )
dv
≥ 0, so −αL2 drdvL −
αS2 drdvS ≤ 0, and so Equation (49) is negative. If G00 (δ) < 0, however, the sign of the equation
is ambiguous.

The constrained equilibrium

To calculate how rS is affected by v, we take derivative of ΓS at the equilibrium values and

obtain

∂ΓS ∂ΓS drS

0= + . (50)
∂v ∂rS dv

When X is sufficiently large, the term X[(f − s)G00 (rS − s + v) − G0 (rS − s + v)] dominates
∂ΓS ∂ΓS
∂v
. Since −G0 (δ)/f < G00 (δ) < G0 (δ)/f , we know that ∂v
< 0. The second-order condition
drS
implies that ∂ΓS /∂rS < 0. Hence dv
< 0, i.e., rS is decreasing in v.
For deposit market share αS = G(rS − s + v), when X becomes sufficiently large, we
have

drS ∂ΓS ∂ΓS (f − rS )G00 (rS − s + v) − G0 (rS − s + v)

=− / →− . (51)
dv ∂v ∂rS (f − rS )G00 (rS − s + v) − 2G0 (rS − s + v)

37
Hence,

d(rS − s + v) drS −G0 (rS − s + v)

= +1= , (52)
dv dv (f − rS )G00 (rS − s + v) − 2G0 (rs − s + v)

d(rS −s+v)
where the numerator and the denominator are both negative. So dv
> 0, i.e., αS is
increasing in v and αL is decreasing in v.
For weighted average deposit rate, take derivative with respect to v:

d dαS drS
(αL s + αS rS ) = (rS − s) + αS
dv dv dv
(f − s)drS /dv + rS − s
= αS (53)
f − rS

αS
where the second equality uses dαS /dv → (f −rS )
(drS /dv + 1), implied by the small bank’s
FOC when X is sufficiently large. The derivative is negative if and only if (f −s)drS /dv+rS −
00 0
−rS )G (y)−G (y)
s < 0. Write y = rS −rL +v. Plugging in drS
dv
= − (f(f−r 00 0
S )G (y)−2G (y)
, the derivative is negative if
f +s−2rS 0
and only if G00 (rS −s+v) ≤ (f −rS )2
G (rS −s+v). We now show that f +s−2rS ≥ 0. We know
that rS is decreasing in v. So we only need to show, given s, f + s − 2rS ≥ 0 when v = 0. Let
dl(x)
l(x) = (f −x)G0 (x−s)−G(x−s), then l(rS ) = 0, and dx
= (f −x)G00 (x−s)−2G0 (x−s) < 0
under the condition that −G0 (δ)/f < G00 (δ) < G0 (δ)/f for any δ ∈ [0, f − s + v]. To show
f −s 0 f −s
that f + s − 2rS ≥ 0, we only need l( 21 (f + s)) ≤ 0. That is, 2
G( 2 ) − G( f −s
2
) ≤ 0.
dm(x)
This is true because if we let m(x) = xG0 (x) − G(x), then dx
= xG00 (x) ≤ 0. And since
m(0) = 0, we have m( f −s
2
) ≤ 0.
For loan quality thresholds and individual banks’ loan volumes, the same proofs for the
unconstrained equilibrium apply and are omitted.
1 + f αL2 (f − s) + αS2 (f − rS )
Total loan volume equals αL (1 − qL∗ ) + αS (1 − qS∗ ) = 1 − + .
A A
Its derivative with respect to v is

 
1 dαS 2 drS
[2αS (f − rS ) − 2αL (f − s)] − αS . (54)
A dv dv

38
Its sign is ambiguous because while the first term in the brackets is negative, the second
term is positive.

Proof of Proposition 5

In the unconstrained equilibrium, the difference of the two banks’ FOCs leads to Equation
1
α α (f −rL )(f −rS )G0 (rS −rL +v)
A L S
(16) . So rS − rL does not vary with f . Further, let B ≡ X+αL (1−qL ∗ )+α (1−q ∗ )
S
, then
S

the two FOCs are separately written as

(f − rL )G0 (rS − rL + v) = 1 − G(rS − rL + v) + B (55)

(f − rS )G0 (rS − rL + v) = G(rS − rL + v) + B (56)

When X is sufficiently large, B → 0. Hence as f changes, rL and rS move one-for-one with

f.
In the constrained equilibrium, let ΓS = dΠS /drS . To study the sensitivity of rS to f ,
we take derivative of ΓS at the equilibrium values and obtain

∂ΓS ∂ΓS drS

0= + (57)
∂f ∂rS df

When X becomes sufficiently large, we have

drS ∂ΓS ∂ΓS 1

=− / → G00 (r −s+v) (58)
df ∂f ∂rS −(f − rS ) G0 (r S
+2
S −s+v)

We know that (f − rS ) is decreasing in s and increasing in v, and that rS − s + v is decreasing

G00 (δ)
in s and increasing in v. If G satisfies that G0 (δ)
is increasing in δ for any δ ∈ [0, f − s + v],
then the denominator is increasing in s and decreasing in v . Also, since −G0 (δ)/f < G00 (δ) <
drS
G0 (δ)/f for any δ ∈ [0, f − s + v], the denominator is positive. So df
is decreasing in s and
increasing in v.
Given v and s, we solve for the cut-off value f ∗ that separates the constrained equilibrium

39
and the unconstrained equilibrium. Let rS∗ be the equilibrium value of rS when f = f ∗ . When
X is sufficiently large, the small bank’s FOC and the large bank’s FOC are

(f ∗ − rS∗ )G0 (rS∗ − s + v) − G(rS∗ − s + v) = 0 (59)

(f ∗ − s)G0 (rS∗ − s + v) − 1 + G(rS∗ − s + v) = 0 (60)

Taking the difference of the two equations, we have

∆ = (rS∗ − s)G0 (rS∗ − s + v) − 1 + 2G(rS∗ − s + v) = 0 (61)

So rS∗ − s = m(v) is a function of v. By the small bank’s FOC, we have

G(m(v) + v)
f∗ = s + + m(v) (62)
G0 (m(v) + v)

Hence f ∗ increases one-for-one with s. To show that f ∗ is decreasing in v, we take derivative

dm(v)
of ∆ to solve for dv

∂∆ ∂∆ dm(v)
0= + (63)
∂v ∂m(v) dv

dm(v) 00 0
Hence, dv
= − ∂∆ / ∂∆ = − m(v)G
∂v ∂m(v)
(m(v)+v)+2G (m(v)+v)
m(v)G00 (m(v)+v)+3G0 (m(v)+v)
. Denoting y = m(v)+v = rS∗ −s+v,
we have

df ∗
 
d G(y) dm(v)
= +
dv dv G0 (y) dv
G (y) [−m(v)G (y) − G0 (y)] − G00 (y)G0 (y)G(y)
0 2 00
= (64)
G0 (y)2 [m(v)G00 (y) + 3G0 (y)]

where rS∗ − s = m(v) > 0. For G that satisfies 0 ≤ G00 (δ) < G0 (δ)/f for any δ ∈ [0, f − s + v],
df ∗
we know the denominator is positive, and the numerator is negative, so dv
< 0.

40
Appendix B: Fitting U.S. deposit rates

In this appendix, we illustrate the predictive performance of our model using U.S. data on
deposit rates from 1986 to 2021. While our model is highly stylized and features only two
banks, it captures qualitative features of deposit rates: they are lower than and somewhat
non-responsive to the interest rate on reserves.
We use the model to predict deposit rates. The opportunity cost of funds for banks is
determined in part by either the IOR rate or the federal funds rate, whichever is larger.
In the period before the 2008-09 crisis the relevant rate was the federal funds rate. In the
period after the crisis the relevant rate was (generally) the IOR rate. We use the higher of
the two rates in each period as f , and apply the model under a specific parameterization.
Specifically, we assume the convenience value for deposit at the large bank, δ, is distributed
uniformly from 0 to 3.5%. The resulting predicted data series seems to fit the actual data
reasonably well. Figure 4 shows the actual and predicted U.S. deposit rates from 1986Q1
to 2008Q2 relative to the federal fund rate. Figure 5 shows the actual and predicted U.S.
deposit rates from May 2009 to February 2021 relative to the IOR rate.
In Figure 4, predicted deposit rates match the levels of actual deposit rates quite well
and move almost one-for-one with the federal funds rate. The main deviation from the actual
data is that predicted rates are too sensitive to changes in the federal funds rate. This is not
surprising, as in our model, deposit rates of both the large and small bank move one-for-one
with the federal funds rate in the unconstrained equilibrium. A noticeable deviation occurs
during the pre-crisis period from 2001Q3 to 2004Q3. This is when the large bank’s predicted
deposit rate is constrained at the zero lower bound. The transition to low deposit rates
associated with the constrained equilibrium is immediate in our model, but not in the data.
Our model does not build in any “stickiness” into the deposit rate that would be necessary
to match the data more closely.

41
 0.1
Fed Fund effective rate
0.09 Actual deposit interest rate
Model predicted deposit interest rate
0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0
1985 1990 1995 2000 2005 2010

Figure 4: Actual and predicted U.S. deposit rates from 1986Q1 to 2008Q2. Domestic deposit
rates are quarterly, calculated from call reports, as total interest expense on domestic deposits
divided by total domestic deposits, multiplied by 4. The model-implied interest rate is the
weighted average of the large bank’s and the small bank’s deposit rates, weighted by their
market shares. Model parameters: G(δ) = δ/0.035, A = 1.5, X = 10, s = 0.

0.025
Interest on excess reserve
Actual deposit interest rate
Model predicted deposit interest rate
0.02

0.015

0.01

0.005

0
2008 2010 2012 2014 2016 2018 2020 2022

Figure 5: Actual and predicted U.S. deposit rates from May 18, 2009 to February 1, 2021.
Weekly deposit rates for amounts less than $100,000 are obtained from the FDIC through
FRED. The model-implied interest rate is the weighted average of the large bank’s and the
small bank’s deposit rates, weighted by their market shares. Model parameters: G(δ) =
δ/0.035, A = 1.5, X = 10, s = 0.

The time period in Figure 5 is characterized by a long stretch of near zero rates in the
Federal Funds market and an IOR rate of 25 basis points. As the IOR rate was typically

42
higher than the federal funds rate, the IOR rate is the relevant variable for predicting deposit
rates. Deposit rates fell slowly during this period toward zero until the Fed began to raise
the IOR rate in December 2015. The Fed raised the IOR rate multiple time reaching a peak
of 2.40% from December 2018 to April 2019, but deposit rates reacted very slowly. Our
model’s predicted deposit rate captures this non-responsiveness. It is still too sensitive to
changes in IOR compared to the data, but the deviation is not large. The low deposit rates
that are predicted by our model occur because at the low IOR rates that existed during most
of this period the zero lower bound is binding in our model and hence average market rates
are determined largely by large bank deposit rates which are constrained at zero.

References

Agur, I., A. Ari, and G. Dell’Ariccia. 2022. Designing central bank digital currencies. Journal
of Monetary Economics 125:62–79.

Andolfatto, D. 2021. Assessing the impact of central bank digital currency on private banks.
The Economic Journal 131:525–40.

Bank of England. 2020. Central bank digital currency: Opportunities, challenges and design.

Bank of England Quarterly Bulletin. 2014. Money creation in the modern economy.

BIS. 2020. Central bank digital currencies: Foundational principles and core features.

———. 2021. Gaining momentum – results of the 2021 BIS survey on central bank digital
currencies.

Brunnermeier, M. K., and D. Niepelt. 2019. On the equivalence of private and public money.
Journal of Monetary Economics 106:27–41.

Chiu, J., M. Davoodalhosseini, J. H. Jiang, and Y. Zhu. 2019. Bank market power and

43
 central bank digital currency: Theory and quantitative assessment. Working Paper, Bank
of Canada.

Diamond, D. W., and P. H. Dybvig. 1983. Bank runs, deposit insurance, and liquidity.
Journal of Political Economy 91:401–19.

Drechsler, I., A. Savov, and P. Schnabl. 2017. The deposits channel of monetary policy. The
Quarterly Journal of Economics 132:1819–76.

Duffie, D., and A. Krishnamurthy. 2016. Passthrough efficiency in the fed’s new monetary
policy setting. Jackson Hole Symposium, Federal Reserve Bank of Kansas.

Fernández-Villaverde, J., D. Sanches, L. Schilling, and H. Uhlig. 2021. Central bank digital
currency: Central banking for all? Review of Economic Dynamics 41:225–42.

Frost, J., L. Logan, A. Martin, P. E. McCabe, F. M. Natalucci, and J. Remache. 2015.

Overnight RRP operations as a monetary policy tool: Some design considerations. Federal
Reserve Bank of New York Staff Report, No. 712.

Garratt, R., and M. J. Lee. 2021. Monetizing privacy. Federal Reserve Bank of New York
Staff Report, No. 958.

Honka, E., A. Hortaçsu, and M. A. Vitorino. 2017. Advertising, consumer awareness, and
choice: Evidence from the U.S. banking industry. The RAND Journal of Economics
48:611–46.

Keister, T., and D. R. Sanches. 2021. Should central banks issue digital currency? Working
Paper, Federal Reserve Bank of Philadelphia.

Lagarde, C. 2018. Winds of change: The case for new digital currency. Speech at Singapore
Fintech Festival.

Li, Y., and Y. Li. 2021. Payment risk and bank lending. Fisher College of Business Working
Paper No. 2021-03-017.

44
Martin, A., J. McAndrews, and D. Skeie. 2016. Bank lending in times of large bank reserves.
International Journal of Central Banking 12:193–222.

Parlour, C. A., U. Rajan, and J. Walden. 2022. Payment system externalities. Journal of
Finance 77:1019–53.

Piazzesi, M., and M. Schneider. 2020. Credit lines, bank deposits or CBDC? Competition
& efficiency in modern payment systems. Working Paper, Stanford University.

45
Previous volumes in this series
1045 The term structure of carbon premia Dora Xia and Omar Zulaica
October 2022

1044 Gender diversity in bank boardrooms and Leonardo Gambacorta, Livia

October 2022 green lending: evidence from euro area Pancotto, Alessio Reghezza and
credit register data Martina Spaggiari
1043 The scarring effects of deep contractions David Aikman, Mathias Drehmann,
October 2022 Mikael Juselius and Xiaochuan Xing

1042 Emerging market bond flows and exchange Peter Hördahl and Giorgio Valente
September 2022 rate returns

1041 The impact of fintech lending on credit access Giulio Cornelli, Jon Frost, Leonardo
September 2022 for U.S. small businesses Gambacorta and Julapa Jagtiani

1040 Quantifying the role of interest rates, Emanuel Kohlscheen

September 2022 the Dollar and Covid in oil prices

1039 Cyber risk in central banking Sebastian Doerr, Leonardo

September 2022 Gambacorta, Thomas Leach,
Bertrand Legros and David Whyte
1038 Building portfolios of sovereign securities Gong Cheng, Eric Jondeau and
September 2022 with decreasing carbon footprints Benôit Mojon

1037 Big Techs vs Banks Leonardo Gambacorta,

August 2022 Fahad Khalil and Bruno M Parigi

1036 The scarring effects of deep contractions David Aikman, Mathias Drehmann,
August 2022 Mikael Juselius and Xiaochuan Xing

1035 Cross-border financial centres Pamela Pogliani and Philip

July 2022 Wooldridge

1034 Debt sustainability and monetary policy: the Enrique Alberola, Gong Cheng,
July 2022 case of ECB asset purchases Andrea Consiglio and
Stavros A Zenios
1033 The Holt-Winters filter and the one-sided HP Rodrigo Alfaro and Mathias
July 2022 filter: A close correspondence Drehmann

1032 Capital flows and monetary policy trade-offs Paolo Cavallino and Boris Hofmann
July 2022 in emerging market economies

All volumes are available on our website www.bis.org.